Recent analyses have reported catastrophic global declines in vertebrate populations^1,2. However, the distillation of many trends into a global mean index obscures the variation that can inform conservation measures and can be sensitive to analytical decisions. For example, previous analyses have estimated a mean vertebrate decline of more than 50% since 1970 (Living Planet Index²). Here we show, however, that this estimate is driven by less than 3% of vertebrate populations; if these extremely declining populations are excluded, the global trend switches to an increase. The sensitivity of global mean trends to outliers suggests that more informative indices are needed. We propose an alternative approach, which identifies clusters of extreme decline (or increase) that differ statistically from the majority of population trends. We show that, of taxonomic-geographic systems in the Living Planet Index, 16 systems contain clusters of extreme decline (comprising around 1% of populations; these extreme declines occur disproportionately in larger animals) and 7 contain extreme increases (around 0.4% of populations). The remaining 98.6% of populations across all systems showed no mean global trend. However, when analysed separately, three systems were declining strongly with high certainty (all in the Indo-Pacific region) and seven were declining strongly but with less certainty (mostly reptile and amphibian groups). Accounting for extreme clusters fundamentally alters the interpretation of global vertebrate trends and should be used to help to prioritize conservation efforts.